cd ~cygnss-deployment
# download CyGNSS data
python API.py
# download ERA5 data and annotate CyGNSS data with wind speed labels
# preprocss (filter) to create hdf5
python Preprocessing.py
# Inference
PYTHONPATH="./externals/gfz_cygnss/":${PYTHONPATH}
export PYTHONPATH
python ./externals/gfz_cygnss/gfz_202003/training/cygnssnet.py --load-model-path ./externals/gfz_cygnss/trained_models/ygambdos_yykDM.ckpt --data ./dev_data --save-y-true --prediction-output-path ./prediction/current_predictions.h5
Create conda
environment using
conda env create --file docker/kernel-env-cuda11.yaml
conda activate cygnss-d
# some packages were not installed correctly
conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch
conda install pytorch-lightning -c conda-forge
pip install global-land-mask
Create Jupyterhub kernel from this environment following https://docs.dkrz.de/doc/software%26services/jupyterhub/kernels.html
- Retrieve user ID and create
.netrc
as described in ... - change the persmission of the file: chmod og-rwx ~/.netrc
Retrieve user ID and API key and create cdsapi
as described in ...